AI-Driven Triage to Enhance Chronic Sinusitis Management: A Quality Improvement Initiative

Short Running Title: AI-Driven Triage for Chronic Sinusitis

Authors:
Bradford G. Bichey MD MPH (Corresponding Author) (ORCID ID: 0009-0006-7293-666X)
Andrew R. Bichey II, BA, Juan V. Rangel

Abstract

Rationale:

Surgical triage for elective chronic sinusitis management often suffers from workflow inefficiencies, prolonged wait times, and administrative burdens. This quality improvement initiative addresses the inefficiencies in traditional pathways of treatment for chronic sinusitis management by assessing an Artificial Intelligence (AI) and Machine Learning (ML) based process to enhance patient care.

Aims and Objectives:

At 100 clinical sites, two-thirds of care coordination efforts experienced scheduling breakdowns, delaying treatment and exacerbating provider burnout. We implemented a commercially available AI-enabled SaaS platform integrating multimodal communication, machine learning algorithms,and automated scheduling with human oversight, to improve patient care, scheduling efficiency, and overall time to treat.

Methods:

A retrospective Quality Improvement Initiative analyzed 10,124 de-identified patient engagements into surgical treatment through the platform. Key measures included time to treatment, prequalification scoring, scheduling delay, patient engagement metrics, ICD-10 predictive accuracy, and resource utilization. No PHI was processed and IRB review was not required.

Results:

• Prequalification & Scheduling: 55% of patients were correctly identified as surgical candidates, reducing average time to treat by 40%.
• Engagement: SMS workflows increased patient engagement by 65%.
• Predictive Accuracy: ICD-10 coding algorithms achieved 82.4% accuracy for ICD-10 code J32.9 (chronic sinusitis, unspecified) and 74.2% for related symptom clusters.
• Efficiency Gains: Overall cost and administrative effort decreased by 50%, with 90% of cases meeting prior authorization criteria and not needing appeal or peer review.

Conclusions:

Deploying an AI-driven SaaS triage system can substantially streamline chronic sinusitis surgical pathways, improve patient engagement, and enhance resource utilization. These findings support broader application of intelligent platforms in surgical care coordination.

Keywords

Chronic Sinusitis; Quality Improvement; Artificial Intelligence; Surgical Triage; Patient Engagement; Software as a Service (SaaS)

Introduction

Problem Description

Chronic rhinosinusitis affects an estimated 12 percent of the U.S. population and is a leading cause of otolaryngology referrals. Delays in surgical evaluation—driven by inefficient scheduling processes, fragmented communication between patients and providers, and administrative bottlenecks—often lead to worsened symptoms, reduced quality of life, and increased healthcare utilization. Since 2019, the U.S. healthcare system has faced increasing challenges in surgical care, impacting patients, practices, and healthcare institutions. Approximately two-thirds of care coordination efforts have efficiency breakdowns, leading to mis-scheduled care and complications with insurance processes. The continued worsening of a healthcare economy with growing administrative burdens and workforce shortages has exacerbated these challenges in the United States.¹ In our network of 100 outpatient clinics, preliminary review showed mean time from initial patient engagement to surgical consultation exceeding 60 days, with two-thirds of cases requiring multiple phone calls or manual reminder workflows to complete scheduling.

Available Knowledge

Prior quality improvement efforts in surgical triage have focused on protocol standardization and care navigator roles, yielding moderate gains in throughput but at high labor cost. Consumer behavior has rapidly shifted to a focus on convenience with a lagging response at the provider level.² Workforce issues and patient behavior are exacerbating financial pressures across the industry causing surgical healthcare to undergo significant cost constraints. Surgical expenditures, which were $572 billion in 2005, are projected to rise to $912 billion by 2025, amounting to 7.3% of the GDP. Per capita healthcare spending is anticipated to reach $8,832 by 2025, representing a significant economic burden for the country.³ Macroeconomic and structural reform attempts have also been insufficient, often overlooking the nuances of surgical care and the diverse needs of patients and providers. More recently, artificial intelligence (AI) applications—especially those leveraging natural language processing and predictive coding—have demonstrated promise in automating preliminary case review, improving diagnostic coding accuracy, and triaging high-risk patients.^4,5,6 Similarly, SMS-based patient engagement platforms have been shown to increase appointment adherence by up to 50 percent in chronic disease populations.⁷ However, few studies have evaluated fully integrated, AI-driven SaaS solutions within a real-world otolaryngology surgical pipeline.

Rationale

We hypothesized that combining machine-learning–enabled prequalification algorithms with automated, bidirectional SMS workflows and streamlined human oversight could address both diagnostic uncertainty and communication gaps. By surfacing likely surgical candidates early—concurrently reducing administrative burden via intelligent automation—this approach should shorten the engagement-to-treatment interval, enhance patient care, and optimize resource utilization (Figure 1).

Specific Aims

• Measure the impact of an AI-driven SaaS triage system on time from initial patient engagement to surgical intervention.
• Assess patient engagement through SMS workflow metrics (response rate, completion of SMS prequalification messaging, and triage into scheduling).
• Evaluate the diagnostic accuracy of ICD-10 predictive models for chronic sinusitis (e.g., J32.9) and related symptom clusters.
• Quantify administrative efficiency gains, including reduction in manual scheduling steps, prior-authorization denials, need for appeals and /or peer-to-peers.

Methods

Context

This Quality Improvement Initiative was conducted across a network of 100 outpatient otolaryngology clinics in the United States. The organization employs a centralized care coordination team using humans-in-the-loop (HITL) AI processes to achieve patient engagement, prequalification, insurance verification, and surgical scheduling (Figure 2). Prior to the intervention, engagement-to-scheduling workflows relied on manual phone calls, email follow-ups, and heterogeneous local EMR processes, resulting in substantial delays and variability in patient experience.

Intervention(s)

We deployed a cloud-based, AI-driven HITL SaaS platform that integrates three core components:

1. Machine-learning prequalification: A predictive model trained on historical patient engagement data and coded diagnosis data to assign a “surgical candidacy” score (0–1) for chronic sinusitis (ICD-10 J32.9 and related codes).
2. Bidirectional SMS workflow: Automated text messages prompting patients to complete conversational discussion about their symptoms and prior treatment as well as validated symptom questioning (SNOT-22), confirm contact information and insurance status, and schedule appointments via embedded links.
3. Human oversight dashboard: A web interface for care coordinators to review algorithm flags, override decisions, and manage exceptions (e.g., urgent flags, language needs) .

The platform was rolled out in a staggered deployment from January through May 2023, with incremental site onboarding and standardized training sessions for patient concierges.

Study of the Intervention(s)

We performed a retrospective, pre–post analysis of de-identified patient engagement records from June 2023–December 2024. All patient journeys for adult patients (≥18 years) with a primary diagnosis of chronic sinusitis were included. Key phases—patient engagement, prequalification, scheduling completion, and surgical consultation—were logged automatically. Care coordinators documented overrides and exceptions in the dashboard.

Measures

Primary outcome: Time (days) from initial patient engagement to completed surgical intervention.
Secondary outcomes:

• Prequalification rate: Proportion of patient engagements flagged as surgical candidates (score ≥0.7).
• SMS engagement metrics: Response rate to initial outreach, timing and completion rate of necessary steps into scheduling.
• Predictive accuracy: Sensitivity, specificity, and overall accuracy of the prequalification model against final clinical determination.
• Administrative efficiency: Decrease in number of front desk/scheduling personnel needed, schedules completed outside the system, and rates of prior-authorization denials, appeals, and peer-to-peers.

Analysis

Descriptive statistics (means ± SD, medians with interquartile ranges) characterized baseline and post-intervention cohorts. We compared continuous outcomes using two-sample t-tests or Wilcoxon rank-sum tests (as appropriate) and categorical outcomes with chi-square tests. Model performance metrics were computed with 95% confidence intervals. All analyses were conducted in R version 4.2.1. A p-value < 0.05 denoted statistical significance.

Ethical Considerations

The initiative used only de-identified data extracted from operational records; no protected health information was processed. The project met institutional criteria for Quality Improvement and did not require formal IRB review. All platform vendors complied with HIPAA and industry-standard data security protocols.

Results

Implementation and Cohort

Between January and May 2023, our AI-driven triage platform was rolled out in a staggered fashion across 100 outpatient ENT clinics. The final analytic dataset comprised 10,124 de-identified patient journeys for chronic sinusitis treatment, each with timestamps for initiation of patient engagement, prequalification, scheduling completion, and surgical consultation.

Primary Outcome: Engagement-to-Surgical Intervention Interval

After full deployment, the mean time from initial patient engagement to completed surgical intervention decreased by 40%, from a baseline of 60 days to 36 days (p < 0.001). This acceleration of the care pathway was accompanied by a reduction in variability, with the interquartile range narrowing from 30 days to 18 days.

Secondary Outcomes

Prequalification Rate & Predictive Accuracy

• 55% of journeys were flagged as surgical candidates (score ≥ 0.7) and scheduled for procedure evaluation.
• For ICD-10 code J32.9 (unspecified chronic sinusitis and related airway codes), the predictive model achieved 82.4% accuracy; for symptom cluster codes (e.g., R68.89, R51, R9.82) 74.2% accuracy; and for H/I codes (e.g., sick sinus syndrome, otalgia) 70.6% accuracy.

Patient Engagement

Automated SMS outreach led to a 65% increase in patient engagement—measured as response to initial text invite into SMS engagement and completion of the necessary steps into scheduling—compared to historical phone-based workflows.

Clinical Severity Metrics

A subset of 198 patients with both CT and SNOT-22 data was analyzed for quality. In this subset, the mean Lund-Mackay CT score was 15.1, mean chronic sinsusitis symptom duration 13.4 months, and mean SNOT-22 score 57.6, indicating moderate to severe disease in those predicted to have surgical disease prior to scheduling and progressing to consultation (Figure 3).

Administrative Efficiency & Cost

• Manual scheduling steps per patient dropped by 50%.
• 90% of surgical candidates met insurance prior-authorization criteria on first submission, reducing labor-intensive appeals and peer-to-peer reviews.
• Overall administrative costs were halved, with an average reduction across office front desk and scheduling staff from 2.3 to 1.2. A single patient concierge managed over 10,000 consults across 24 months—a testament to the platform’s scalability.

Contextual and Unintended Findings

• Insurance Distribution: Providers configured the system to prioritize commercial payers; Blue Cross Blue Shield accounted for 43% of consultations, United 24%, Cigna 15%, Medicare 9%, Medicaid 3%, and other insurers 2% (Figure 4).
• Human Oversight: Coordinators used the override dashboard in 12% of cases, most often for language-concordant outreach or urgent flagging, reinforcing the value of a human-in-the-loop model.
• No Significant Data Gaps: Less than 0.1% of patient engagements had missing timestamps or incomplete workflow data; these were excluded from time-to-consult analyses but did not bias overall findings.

Discussion

Summary

In this Quality Improvement Initiative across 100 outpatient ENT clinics, implementation of an AI-driven SaaS triage platform reduced the mean engagement-to-consultation interval by 40% (from 60 to 36 days), increased patient engagement by 65% via automated SMS workflows, and achieved diagnostic coding accuracies of 82.4% for J32.9 and 74.2% for related symptom clusters (Figure 5). Administrative workload was halved, with prior-authorization success on first submission rising to 90% and manual scheduling steps reduced by 50%.

Interpretation

These findings demonstrate that integrating AI/ML prequalification with bidirectional SMS and human oversight can streamline surgical pathways in chronic sinusitis. The reduction in wait times likely reflects both timely identification of surgical candidates prior to scheduling and elimination of repetitive manual coordination tasks. This aligns with prior reports of AI-assisted diagnostic support and digital outreach enhancing chronic disease management.⁸ Notably, the human-in-the-loop model—evidenced by a 12% override rate—ensured that algorithmic decisions were contextualized by the patient concierge, preserving patient safety and equity, particularly for language or socioeconomically driven exceptions.

These findings reflect the software’s nuanced understanding of symptomatology and medical coding, which is not feasible in many practices at the front desk or with phone based call centers.⁹ As an additional benefit, by analyzing real journey interactions, the software evolved to optimize the prequalification, triage, and scheduling of future journeys, enhancing the operational efficiency of the healthcare provider’s practice.

Previous research has shown that institutions can consider automated texting strategies to augment healthcare support for patients with limited added burden to staff.¹⁰ By combining the advantages of SMS with HITL AI we aimed to decrease total time to surgical treatment as well as achieve secondary benefits that improved efficiency and cost savings across the system.

Figure 6 illustrates a head-to-head comparison of three surgical triage methods—(1) SMS-and AI-driven, human-in-the-loop (HITL) processes; (2) conventional phone-based workflows; and (3) reliance on expert triage staff alone—across four domains: patient acquisition, optimized scheduling, cost-savings efficiency, and overall expense. We observed that the SMS/AI/HITL column is uniformly rated “Efficient” for acquisition, scheduling, and cost savings, and “Low Cost” for expense, reflecting the dramatic reductions in manual effort and operating overhead. By contrast, both phone-based and expert-staff methods score “Not Efficient” on all three performance metrics and incur “High Cost,” underscoring persistent bottlenecks (multiple callbacks, scheduler backlog) and the labor-intensive nature of expert-only review. Taken together, this matrix visually reinforces that only an integrated digital platform can simultaneously accelerate throughput, maintain accuracy via human oversight, and minimize administrative expense—key goals of any scalable surgical quality improvement intervention.

Limitations

Several factors may limit generalizability. First, this was a retrospective, single SaaS solution study without randomized controls; secular trends or concurrent initiatives could contribute to observed improvements. Second, although data was de-identified, we did not capture granular patient-level socioeconomic or comorbidity variables, which may influence both engagement and treatment timelines. Third, while the platform was HIPAA-compliant, we did not formally assess metrics on patient preference or perceptions of digital communication. Finally, early adopter sites may have more experienced patient handling or better local infrastructure, potentially biasing results compared to later-onboarded clinics.

Conclusions

Deploying an AI-enabled SaaS triage system within a human-in-the-loop framework significantly enhanced efficiency, accuracy, and engagement in chronic sinusitis surgical pathways. These results support broader adoption of intelligent automation platforms to reduce care delays and administrative burden in surgical specialties. Future work should include prospective, multicenter trials, incorporate patient-reported outcomes on digital engagement, and evaluate long-term clinical and cost impacts.

---

Conflicts of Interest

All authors are named applicants on U.S. Patent No.18352021 (“Multi-Modal Digital Communication Architecture for Patient Engagement”), Application No. 18/352,021, filed July 14, 2023; published January 16, 2025. In addition, Dr. Brad Bichey serves as Founder and Chief Executive Officer of Nemedic, Inc., the developer of the AI-driven triage platform evaluated in this report. No other competing financial or non-financial interests are declared.

Acknowledgments

The authors thank the concierge care-coordination team for their engagement in platform deployment and data collection; the software development group for iterative feedback on human-in-the-loop intervention; and the patients whose anonymized data made this analysis possible. This Quality Improvement Initiative and manuscript development were supported internally by Nemedic, Inc. No external funding sources were used.

Ethical Approval and Consent to Participate

This work constituted a Quality Improvement Initiative using only de-identified operational data. It met institutional criteria for QI activities and was determined to be exempt from formal IRB review. No PHI was used and patients opted in via the software’s Terms of Use. Ethical concerns such as equity, AI bias, and patient autonomy were considered in system design.

---

References:

1. Mehta A, Awuah WA, Ng JC, Kundu M, Yarlagadda R, Sen M, Nansubuga EP, Abdul-Rahman T, Hasan MH. Elective surgeries during and after the COVID-19 pandemic: case burden and physician shortage concerns. Ann Med Surg (Lond). 2022 Sep;81:104395. doi:10.1016/j.amsu.2022.104395. PMCID: PMC9388274.
2. Das D, Sarkar A, Debroy A. Impact of COVID-19 on changing consumer behavior: lessons from an emerging economy. Int J Consum Stud. 2022 May;46(3):692–715. doi:10.1111/ijcs.12786. PMCID: PMC9111418.
3. Muñoz E, Muñoz W 3rd, Wise L. National and surgical health care expenditures, 2005-2025. Ann Surg. 2010 Feb;251(2):195–200. doi:10.1097/SLA.0b013e3181cbcc9a.
4. Electronic Health Record effects on work-life balance and burnout within the I3 Population Collaborative. J Grad Med Educ. 2017 Aug;9(4):479–484. doi:10.4300/JGME-D-16-00123.1. PMCID: PMC5559244.
5. Nijor S, Rallis G, Lad N, Gokcen E. Patient safety issues from information overload in electronic medical records. J Patient Saf. 2022 Sep;18(6):e999–e1003. doi:10.1097/PTS.0000000000001003. PMCID: PMC9422765.
6. Corder JC. Streamlining the insurance prior authorization debacle. Mo Med. 2018 Jul-Aug;115(4):312–314. PMID: 30228750.
7. Kershaw K, Martelly L, Stevens C, McInnes DK, Silverman A, Byrne T, Aycinena D, Sabin LL, Garvin LA, Vimalananda VG, Hass R. Text messaging to increase patient engagement in a large health care for the homeless clinic: results of a randomized pilot study. Digit Health. 2022 Oct 9;8:20552076221129729. doi:10.1177/20552076221129729. PMID:36238754; PMCID:PMC9551340.
8. Gala D, Behl H, Shah M, Makaryus AN. The Role of Artificial Intelligence in Improving Patient Outcomes and Future of Healthcare Delivery in Cardiology: A Narrative Review of the Literature. Healthcare (Basel). 2024 Feb 16;12(4):481. doi:10.3390/healthcare12040481. PMID:38391856; PMCID:PMC10887513.
9. Griffith KN, Li D, Davies ML, Pizer SD, Prentice JC. Call center performance affects patient perceptions of access and satisfaction. Am J Manag Care. 2019 Sep 1;25(9):e282–e287. PMCID: PMC8177735. NIHMSID: NIHMS1707498. PMID: 31518100.
10. Bressman EB, Long JA, Honig K, Zee J, McGlaughlin N, Jointer C, Asch DA, Burke RE, Morgan AU. Evaluation of an automated text message–based program to reduce use of acute health care resources after hospital discharge. JAMA Netw Open. 2022 Oct 26;5(10):e2238293. doi:10.1001/jamanetworkopen.2022.38293.